Machine learning-based predicting of PCM-integrated building thermal performance: An application under various weather conditions in Morocco

As is now stands, the building sector is among the "big three" significant energy consumer and greenhouse gas contributor in the world. Consequently, there has been a growing movement towards the development and adoption of renewable energy sources, energy efficient measures and energy management strategies. Phase change materials (PCMs) are considered as an alluring option for energy efficiency measures in building. In this paper, a case study building, based on typical residential construction in Morocco, was selected to assess the potential energy savings of adding PCM to the roof for twenty-four cities using the dynamic simulation tool, TRNSYS. Thereafter, thirteen machine learning techniques, including ANN, DT, SVM, ELM, GB, RF, TB, GLRM, GPR, LR, GAM, KRRM and LRR were assessed for predicting the hourly heating, cooling, and total energy consumptions. The models were trained and tested on a dataset that was gathered from the simulations in twenty-four locations in Morocco. The outdoor dry-bulb temperature, the relative humidity, the wind velocity, the wind direction, and the total solar radiation were considered as the key features. The obtained results revealed that using PCM, can effectively lower the total energy demand for all the cities under study, with exception for very cold climate given by Ifrane and Midelt, where the annual total energy consumption shows an increasing trend. Furthermore, comparing the statistical results for each machine learning model, it is found that the SVM model consistently outperforms all models and can be successfully employed to predict the hourly network's outputs of the building.